Você está na página 1de 10

Int. Fin. Markets, Inst.

and Money 15 (2005) 343–352

Gold as a hedge against the dollar
Forrest Capie a,∗ , Terence C. Mills b,1 , Geoffrey Wood a,2
b

Cass Business School, City University, 106 Bunhill Row, London EC1Y 8TZ, UK Loughborough University, Department of Economics, Loughborough, Leics LE11 3TU, UK Received 24 March 2004; accepted 20 July 2004 Available online 28 June 2005

a

Abstract The extent to which gold has acted as an exchange rate hedge is assessed using weekly data for the last thirty years on the gold price and sterling–dollar and yen–dollar exchange rates. A negative, typically inelastic, relationship is indeed found between gold and these exchange rates, but the strength of this relationship has shifted over time. Thus, although gold has served as a hedge against fluctuations in the foreign exchange value of the dollar, it has only done so to a degree that seems highly dependent on unpredictable political attitudes and events. © 2004 Elsevier B.V. All rights reserved.
JEL classification: C22; F31; F33 Keywords: Gold; Exchange rates; Hedge; GARCH

1. Introduction Charles de Gaulle once said, “There can be no other criterion, no other standard than gold. Yes, gold which never changes, which can be shaped into ingots, bars, coins, which has no nationality and which is externally and universally accepted as the unalterable fiduciary value par excellence”. With these famous words, prompted and perhaps written by his
Corresponding author. Tel.: +44 20 7040 8730; fax: +44 20 7040 8881. E-mail addresses: f.h.capie@city.ac.uk (F. Capie), t.c.mills@lboro.ac.uk (T.C. Mills), g.e.wood@city.ac.uk (G. Wood). 1 Tel.: +44 1509 22703; fax: +44 1509 223910. 2 Tel.: +44 7040 8740; fax: +44 7040 8881. 1042-4431/$ – see front matter © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.intfin.2004.07.002

is of interest. gold has no nationality and is not controlled by governments. for many years over a large part of the world. and depended on the gold supply relative to the demand for it. This sets the scene for the detailed econometric modelling. Looking at the entire period for which data on the gold price are available is. Conversely. That does not mean that gold could no longer be a good store of value or protection against exchange rate change. such behaviour could still be sensible. was no longer in place. and also diverted some of the existing stock to non-monetary uses such as jewellery.344 F. Mill. For. Not only that. however. That. But whether it is or not depends on different forces. for example. to changes in the money supply. In such a situation an automatic stabilising mechanism was in place. the basis of the monetary system. It is useful to explain briefly why gold was inevitably a good hedge when it was the basis of the monetary system. Inst. / Int. and it became a commodity like any other. Having gold as money. the automatic stabilising mechanism. occasionally quite substantial. through changes in gold output and use. Suppose that for some reason the price of goods rose relative to gold. to see how well gold protected against currency fluctuations. working from changes in the relative gold price. In the nineteenth century. This built-in stabilising mechanism has been described in detail by several authors (see. 1984). or as the basis of the monetary system. using data from the last thirty years. fluctuations). Barro. and then to consider why gold may have remained a hedge even when these circumstances changed. 1871. and on when currencies strengthen. if not money. gold was a hedge because it was the basis of the monetary system. a considerable degree of price stability in terms of gold was to be expected. therefore. When gold no longer had that role. Fin. and Money 15 (2005) 343–352 adviser Jaques Rueff. when currencies weaken. divisible. Markets. Capie et al. The behaviour of prices was thus taken outside the control of government and central banks. Section 2 explains the historical background to the extent necessary to justify our choice of data period. well worth exploring the past. Even though gold no longer has any role in the monetary system of any major country. it was a standard which held steady its purchasing power in terms of goods over a very long period of years (although there were short-term. if the price of goods fell there was a rise in the relative price of gold. it is. highly likely to be misleading. as understanding that is crucial to our choice of data period. which is reported in Section 3. the gold standard was the monetary system . Conclusions are drawn in Section 4. people switch to gold. they become more confident about the value of currencies. as de Gaulle pointed out. In determining the extent to which gold acts as an exchange rate hedge. Gold as an exchange rate hedge For a long part of the recorded past. in itself. Hence. meant linking a currency to gold at a fixed price. For many years gold was. was indeed the ultimate standard of value. and it may also be of interest in the future. and thus a stimulus to its production. then that ceased. Crucial to this mechanism was gold being the basis of the monetary system. For there was part way through that data period a very substantial regime change. and. It is durable. and switch from gold. this fall in the relative price of gold reduced incentives to produce gold. 1979. Rockoff. de Gaulle summarised both a long-held view of gold and many of the reasons for that view being held. It depends on whether. 2.

at least in principle. Fin. Accordingly. had left it in 1931 and the United States followed in 1933. but it soon became pretty clear that there was to be no official role for gold in the future. If gold were a perfect internal hedge. The focus of the paper is the latter. why might gold be of interest? The answer is two fold. An early examination of these can be found in Carse et al. Some talk of an official role for gold lingered on. but Britain. because “it was a badge of honour and decency”. Gold. it is useful to consider why any firm (or individual) might be interested in using gold as a hedge. a different matter. it seems appropriate for our statistical work to start in 1971. all the available techniques provide what can be termed a “special hedge”. albeit indirectly. Whether it did so in practice is.. These arrangements came under strain in the late 1960s. Inst. That is the subject of the remainder of this paper. Before setting out our approach and our results. There were areas using silver. The price began to move around significantly towards the end of the 1960s. nominal) price would rise at exactly the same rate and time as the number of units of foreign currency per dollar fell.F.e. social. In the newly designed international monetary system that followed the end of the Second World War there remained a vestige of gold. (1980).00 per oz). Countries joined the gold standard in part. the standard’s predominant member before 1914. With such choice available. when there began to be anticipations of breakdown in the system. history starts in 1971. The second is a hedge against changes in the external purchasing power of the dollar. a range of products have developed which mean that one can in effect buy gold without actually taking possession of the physical commodity. a recent study of hedging dollar risk using derivatives is Allayannis and Ofek (2001). Markets. its dollar (i. For our purpose in this study. The first is a hedge against changes in the internal or domestic purchasing power of the dollar. the US dollar was still expressed in terms of a fixed gold price (US$ 35. in contrast. and there was some limited movement in the gold price in the 1950s and 1960s. / Int. is incapable of protecting against fluctuations of currencies in general. There is after all a wide range of ways to hedge against exchange risk. Second. Some countries held on until 1936.. for only since then can we explore the behaviour of gold as a hedge uninfluenced by expectations of future official actions on gold. at least. Although the system actually turned out to be in effect a dollar standard. But gold was the standard to which most countries aspired. If it were a perfect external hedge. Capie et al. First. The Bretton Woods system collapsed completely with President Nixon’s closing of the gold window in 1971. and the sterling standard (of course linked. then. with the aim of assessing to what extent gold was an external dollar hedge. What is meant by saying that gold was a hedge against the US dollar in this period? It could be a hedge in two senses. The thirty years following the outbreak of the First World War saw the gradual crumbling of the gold standard in the face of unrelenting political. to gold) held sway in much of the British Empire. . and economic forces. its dollar (i. of course. and Money 15 (2005) 343–352 345 that predominated in the developed world. Protection is given lest one currency fluctuated against some other specific currency. nominal) price would rise at the same rate and time as a domestic US price index. to quote Joseph Schumpeter’s words on the reasons behind Austria’s decision to link her currency to gold.e.

Sterling. quoted in US dollars per troy ounce.m. Markets. 1. Inst. but with long interruptions for depreciation (i. The fixing members’ representatives relay the price to their dealing rooms and these are in contact with as many dealers as are interested (or who have interested clients). before declining to around $300 by mid-1982. Given the large range of fluctuations in each of the series. dollar appreciation) during the first half of the 1980s. It was found that first differences were all that was required to eliminate nonstationarity for all three series under investigation. i.m.. and the sterling–dollar and yen–dollar exchange rates. The “gold fix” takes place twice daily in the offices of NM Rothschild in London. Detailed statistical analysis of the series may be found in Mills (2004a). each series xt was transformed to xt = xt − xt-1 . at the discretion of the chairman of the fix. and begin the fix with a ‘trying’ price. The data and econometric modelling The data series used are weekly observations on the price of gold. If more gold is required than is offered. then he declares himself a buyer of 1 tonne). When analysing an individual time series. then the price will be adjusted upwards (and vice versa) until equilibrium is reached. fix. particularly in the last ten years after its exit from the ERM in September 1992. has since held relatively stable. (If one is buying 2 tonnes and another is selling 1 tonne. The five members of the fix meet at 10:30 a.e. Capie et al. which is from 8 January 1971 to 20 February 2004. Fin. may change their order or add to it or cancel it at any time. although it too depreciated dramatically during the first half of the 1980s. The gold price increased rapidly between 1971 and the end of 1974 and from September 1976 to January 1980. The position declared by the dealers is the net position outstanding among all their clients.. Fig. The behaviour of the yen has been that of a general appreciation. such nonstationary can often be eliminated by taking differences.e. The dealers. . logarithms were used throughout the analysis. These market members then declare how much gold they are prepared to buy or sell at that price. e. On very rare occasions the price will be fixed when there is disequilibrium. with the price being around $400 during January and February 2004. These rates were chosen as they are from the two most important foreign exchange markets that were in operation for the complete sample period.346 F. which we denote gt . 1 shows the weekly gold price series. / Int. which is seen to range from about $37 at the beginning of 1971 to a maximum of $835 on 18 January 1980. In fact.1 1 Nonstationary time series of the type being analysed here typically contain time varying means and variances. which we generically denote xt . and price change would be initiated elsewhere only if news came in while London was closed. The gold price is the Friday London 3:00 p. The Zurich and New York markets follow London. At this point the price is fixed. and 3:00 p. analysis was conducted on the first difference of the logarithm of the gold price. The two exchange rate series are also shown in Fig. The fix is in this way entirely open and any market user may participate through his bank. and the first differences of the logarithms of the two exchange rates. a total of 1728 observations. The sterling–dollar and yen–dollar exchange rates were taken as those observed closest to the gold price. or the nearest observation to that.m.g. who are in contact with their clients. London time. Since that time it has traded within a range of approximately $250–$500. trends of some form and changing volatility.. although it had again begun to appreciate during the last couple of months of the sample period. it is the dominant market. and Money 15 (2005) 343–352 3.

01 −0. 8 January 1971–20 February 2004.02 −0.03 k=0 −0. thus suggesting that any dynamic relationship.02 k = −3 −0. / Int. and Money 15 (2005) 343–352 347 Fig. of around −0. g (k) x and g k = −4 Sterling Yen −0. however. Of obvious importance are the highly significant negative contemporaneous correlations.F. thus necessitating the use of first differences in the fitted relationships. See. 1. Table 1 reports the cross-correlations between gt and xt for up to four lags. Thus.05 and above are statistically significant at the 5% level.2.05 k = −1 −0. thus providing statistical confirmation of the hedging properties of gold.025. Inst. there is little evidence of significant cross-correlations between gt and xt .06 −0. The need for first differencing was formally tested via the application of unit root tests. sterling–dollar exchange rate (sterling). negative values of k correspond to x ‘leading’ g. Table 1 Estimated cross-correlations between r x. Fin.01 k=2 −0. The standard error attached to these cross-correlations is of the order of 0.00 k = −2 0. Prior testing found that the logarithm of the gold price was not cointegrated with either of the exchange rates. Markets. When relating one nonstationary series to another. for example. g (k) is the correlation between xt + k and gt . However. between the change in the gold price and the changes in the two exchange rates.05 0. so (absolute) values of 0.02 −0. will be very short-lived. which informally means that the series contain common trend components that a linear combination of the two annihilates.16 k=1 −0.02 −0. end week. if it exists at all.01 0. Gold price in dollars (Gold). Capie et al. This is the situation when the two series are cointegrated. it is possible that differencing may not be required for statistical analysis to be undertaken. Hamilton (1994) and Mills (1999) for detailed discussion of all these concepts. yen–dollar exchange rate (yen).06 k=3 0.07 r x.24 −0.03 −0. .01 k=4 −0.

(0. with innovations.074 (0. A variety of such processes were entertained: the conventional GARCH process. whilst mis-specifying the dynamics of the conditional error variance process (i.508 log(σt2−2 ) + 0. Markets. On defining the conditional 2 error variance as E(ε2 t | gt −1 . The combination of process and innovation distribution that produced the maximum likelihood value was selected. (1) in both cases. ) = σt . assumed to be distributed as either Gaussian. the threshold GARCH process. which we know not to be the case from Fig. Since β0 is the short-run elasticity.310 (0. In both cases the estimate of β0 is negative.188) (0.4 α0 and β1 were found to be insignificant in the ‘mean’ Eq. See Mills (1999. Fin.214 yt + 0. as a non-zero constant would imply either a generally upward or downward drifting process for the levels of the gold price.023) log(σt2 ) = − 0. we consider autoregressive distributed lag models of the form gt = α0 + α1 gt −1 + β0 xt + β1 xt −1 + εt (1) Thus the current change in (the logarithm of) the gold price is assumed to depend linearly on the current and past change in (the logarithm of) the exchange rate and the past change in the gold price itself. Neither finding is surprising.633 log(σt2−1 ) + 0. such conditional variances were then modelled as ARCH processes.080) (0.024) ε t −1 . 1.050) εt −1 σt −1 + 0. . As is now conventional.148) (0.081) (0. confirming that gold is indeed a hedge against exchange rate changes. . Very similar models were selected for both exchange rates ( st and yt are used below to denote the log-changes in sterling and the yen..049 gt −1 + εt (0.024) log(σt2 ) = − 0. / Int. by ignoring its time varying nature) will often manifest itself in a mean process that contains spurious dynamics. . εt . chapter 4) for example. xt −1 . are distributed as Student’s t with v degrees of freedom): Sterling equation gt = − 0.041 gt −1 + εt (0.3 gt = − 0. and Money 15 (2005) 343–352 Since we are particularly interested in the response of the gold price to changes in exchange rates. gt . εt .331 (0. and the exponential GARCH process (EGARCH).348 F.064 . the residuals from OLS estimation of this model showed strong evidence of time varying conditional error variances.342 log(σt2−2 ) + 0. gold is thus inelastic in .314 st + 0. Capie et al. respectively).024) (0.147) (0. Inst. ε ∼ t (v). thus rendering statistical inference problematic. for discussion of the various models entertained here. ε ∼ t (v).047) εt −1 σt −1 ε t −1 + 0. or generalised exponential.443 + 0. σ t −1 v = 4.412 + 0.465 log(σt2−1 ) + 0. In both cases.e.025) σt −1 Yen equation v = 4. Student’s t.027) (0.191) (0. the following models were arrived at (the notation εt ∼ t(v) denotes that the innovations.

as can be seen from the plot of the estimated value of σ t (the conditional error standard deviation) from the sterling equation shown in Fig. 2004b. and is very similar to the short-run elasticity. It stabilised from the mid-1980s onwards.97. and Money 15 (2005) 343–352 349 Fig.214/(1 − 0. The presence of both the lagged standardised error εt − 1 /σ t − 1 and its absolute value implies that innovations have asymmetric effects. provides further analysis of this feature of the data). the short-run with respect to both exchange rates. and the points are seen to fall into four distinct ‘strata’. Volatility is high throughout the 1970s and particularly so during the early months of 1980. an εt − 1 > 0 (‘good’ news: an unanticipated increase in the price of gold) will have a greater impact on the conditional error variance. The long-run response is reached extremely swiftly. so that shocks are extremely persistent. which makes the error distribution highly fat-tailed. but the autumn of 1999 saw an upsurge in volatility that gradually dissipated during the first quarter of 2000. but because the coefficients on both terms are positive. respectively. Capie et al. 2 (the analogous plot from the yen equation is almost identical). 2. These strata delineate four non-overlapping periods. Fig.223/(1 − 0. in well under a week.2) process with Student’s t-distributed innovations.223 and −0. / Int. The degrees of freedom associated with the Student’s t-distribution is of the order of 4. The conditional error variance equation was found to be best modelled as an EGARCH(1. . Fin. The coefficients on the lagged conditional variances in the equation sum to 0.049) = −0. and hence the volatility of gold. Markets. than an εt − 1 < 0 (‘bad’ news: an unanticipated fall in price). 3 shows the scatterplot between gold and sterling. Given the very different behaviour of the gold price pre. a feature of gold price data that is analysed in some detail in Mills (2004a).041) = −0. It is also inelastic in the long-run. scatterplots of the logarithms of the gold price against both of the exchange rates were investigated. identified by different plotting symbols. Inst.F. Conditional error standard deviation from the sterling equation. a consequence of the small estimate of α1 .330 for sterling and the yen. for these elasticities are −0.and post-1982 and the different exchange rate regimes that were in operation between 1971 and 2002. (Mills.

Markets.15 accompanied by a depreciation in sterling from 0.444 to 0. The first of these has the September 1979 end-point. However. 3.25.5 to $123. Inst. The second ends in the third week of October 1976. 4.250–$2. 1971–2004. shown in Fig. The yen scatterplot.607 to 0. 4. while the second ends with the Plaza agreement of September 1985. Capie et al.168).588). and Money 15 (2005) 343–352 Fig. this depreciation occurred in the middle of a five-week period when gold soared from $300. Fig. The first ends in the second week of January 1974. when gold jumped from $115. Fin. / Int.55 to $397. when the gold price jumped from $122 to $130. when sterling depreciated from 0.630 ($1.350 F. Scatterplot of gold against the yen. The third period ends in early September 1979. Scatterplot of gold against sterling. . has just three strata identified.461 ($2. 1971–2004.648–$1.

Fin. This raises two questions: why has it been a hedge? and why did the extent to which it served as a hedge vary as it did? Gold served as a hedge because it is a homogeneous asset unlike. First.302 −0. and Money 15 (2005) 343–352 Table 2 Sub-period estimates Sample period (a) Sterling 1/08/71–2/20/04 1/18/74–10/22/76 10/29/76–9/07/79 9/14/79–2/20/04 (b) Yen 1/08/71–2/20/04 1/22/71–9/07/79 9/14/79–9/20/85 9/27/85–2/20/04 Short-run elasticity −0.223 −0.2) EGARCH(1. simpler models were found to be warranted (no model is reported for the first sterling sub-period from 1971 to 1974 as no significant relationship could be found).1) EGARCH(1. Capie et al.2) EGARCH(1.302 −0. The elasticity estimates and features of the conditional variance processes and error distributions are reported in Table 2.4) G(1. but underlying all these reasons is that gold cannot be produced by the authorities that produce currencies. and therefore is easily traded in a continuously open market. and so are close to those from the final sub-period. Inst. The complete sample elasticities are effectively weighted averages of the sub-period estimates. Such an expectation could well lead the majority of people to ride out the fluctuation rather than .266 −1. property.980 −0. In all cases.569 −0. Models of the type (1) were thus fitted to each of the sub-periods identified by this analysis. particularly so for sterling.1) EGARCH(1.F. Why has the extent to which gold has served as a dollar hedge varied from time to time? There are at least three important influences on this. elasticities are negative and they are inelastic except for the sterling 1976–1979 sub-period.179 Conditional variance process EGARCH(1.2) EGARCH(1.3) t(3. debase its value cannot by similar means debase the value of gold. Responses are always rapid.330 −0. v < 2 signifies a fat-tailed distribution. Markets. G(v) denotes generalised exponential distribution with parameter v.319 −0.788 −0.214 −0.519 −0. some of which have been discussed in this paper. on some occasions there may have been a firm expectation that the exchange-rate fluctuation was temporary.179 Long-run elasticity −0.8) t(4. say.216 −1.1) GARCH(1.2) EGARCH(1. 4. with short and long-run elasticities typically being very close to each other. It acquired the attributes of an asset for a wide variety of reasons. though the extent to which it was so has varied over the last thirty years.1) 351 Error distribution t(4.319 −0. / Int.1) N denotes Gaussian distribution.0) G(1. Conclusions The conclusions are that gold has been a hedge against the dollar.9) N t(4. This sub-period is also unusual in that it is the only one for which the conditional variance process has a simple GARCH structure and the error distribution is Gaussian: in all other cases EGARCH processes with fat-tailed distributions are the norm. although in several cases.314 −0.3) t(5. from time to time. This means that those who can increase the supply of money and therefore.

). Ofek. R. 1984. M. seventh ed.C. Markets. Examples of such variation. References Allayannis.. on some other occasions. hedging. 1994.. Financing Procedures in UK Foreign Trade. Capie et al.. in press. affect the expected future supply of gold. Exchange rate exposure. Inst. 273–296. The results of this study thus both show that gold has served as a hedge against fluctuation in the foreign exchange value of the dollar. Longmans Green and Company. gold serves as a hedge very much faute de mieux.C.J. S. The Econometric Modelling of Financial Time Series. Carse. Secondly. NJ. In: Bordo.. 13–33. Significance 1. Thirdly. (Eds.. official attitudes to gold may vary and the official sector is a significant holder of gold stocks. its Costs of Production and Commodity Prices. more recently. and Money 15 (2005) 343–352 rearrange their portfolios. and the use of foreign currency derivatives. Hamilton. 1871. but that it has done so to a degree that seems highly dependent on somewhat unpredictable political attitudes and events. J. when the Greek central bank sold gold so as to diversify into bonds issued by its own government. Statistical analysis of daily gold price data. Principles of Political Economy. G. J. London. Cambridge. 559–566. Chicago. T. Cambridge.352 F. University of Chicago Press. Some Evidence on the Real Price of Gold. Wood. 2001. Money and the price level under the gold standard.... . Mills. 2004b. private sector attitudes to gold may have been affected by problems in gold producing countries. Cambridge University Press. Fin. of course. E. Barro. G. Such problems would. when official sectors sold gold at a time or for a reason that seemed puzzling to the observer. Cambridge University Press.C.E. Journal of International Money and Finance 20. Physica A 338. 1980. Mills. 1999. Exploring the relationship between gold and the dollar.. A Retrospective on the Classical Gold Standard. Schwartz. Economic Journal 89..G. second ed.. Princeton. Time Series Analysis. 1821–1931. H. Williamson. were when the British government sold at a trough in the market and. T. 1979.. Mills. A.S. Rockoff. 2004a. Mill. T. J. This may suggest that despite the words of de Gaulle quoted at the start of this paper. / Int. Princeton University Press.